Minefield Detection Based On Synthetic Aperture Radar Images

With the development of technology, thousands of landmines are laid in a few minutes within a large area, and all of them are forming minefield at last. That makes the ability of delaying and killing much stronger than before. The action of minefield is redefined in modern military operations. Therefore, the research on minefield detection has both theoretical and practical values. Simultaneously, synthetic aperture radar(SAR) is able to provide a powerful surveillance capability. When SAR works at low frequency, it can penetrate grass or surface, and find the targets which are buried in them. SAR will be developed into a very important tool of dealing minefield without doubt. Thus, this paper focuses on the procedure of minefield detection and flexible methods.Minefield is consisted of many independent landmines, and their distribution is unknown in advance. As a result, the method of minefield detection is different with classical target detection method, and the detection procedure of target with independent structure is not suitable for minefield. For the reason of convenience and improving the applicability of the method, point set target(PST) and element target(ET) are defined in Chapter 1. Generally, ET is an independent target, and PST is consisted of many ETs. It is visible that minefield is PST and landmine is ET. This paper wants to establish a detection frame of PST. For minefield, in terms of the detection frame of PST, the works of this thesis are summarized as follows:Chapter 2 researches the prescreening of minefield elements, namely the prescreening of landmine. For landmine in airborne SAR image, the problems of small areas and weak scattering always exist. These problems lead a high false alarm rate(FAR) in sense at the high detection rate, and the accurate of demarcation is getting worse. Aiming at this problem, the thesis analyzes the applicability of classical algorithm in prescreening for landmine at first; Then, the grid-based clustering algorithm using area feature is designed, which can reduce false alarms and improve computation efficiency as well as simplify detection process. For more detailed morphological feature, in order to keep low FAR in high detection rate, many algorithms are applied to improve the signal-to-clutter ratio, such as extended fractal algorithm and morphological component analysis.Chapter 3 researches the discriminating of minefield elements, namely the discriminating of landmine. Clutters are discarded in the prescreening of landmine. But there still exist many clutters which have the similar morphology with landmine, such as truck, hole and rock. Therefore, more efficient features need to been extracted for discriminating. Aiming at solving this problem, firstly, the paper researches the echo characteristics in time and frequency domain through the scattering model of landmine, and get the double-dumps feature in physical term. Secondly, the parameters of double-dumps are extracted by the design of image sparse decomposition. Thirdly, the information of time-frequency domain is acquired by sparse time-frequency representation, and the sequential aspect features are formed by sampling of azimuth angles in order. Finally, discriminating component is added in feature extraction for making it more efficient and simplifying discriminating process.Chapter 4 researches the extraction and demarcation of minefield. Firstly, suspected minefield is extracted by the algorithms of discarding isolated points and hierarchical clustering, reducing the affection from one another minefield. Secondly, the distribution of minefield, which is laid by rocket bombs, is acquired by a lot of simulation. Thirdly, real minefield is confirmed by the extraction of distribution parameters of suspected minefield. Finally, in order to accomplish the demarcation of discriminated minefield, marginal landmines are extracted.Chapter 5 makes a summary of the whole thesis, and gives the research prospect in future.